Volume 38, Issue 22
Free Access

Observational changes and trends in northeast Pacific wave records

Johannes Gemmrich

Johannes Gemmrich

Department of Physics and Astronomy, University of Victoria, Victoria, British Columbia, Canada

Search for more papers by this author
Bridget Thomas

Bridget Thomas

Climate Research Division, Science and Technology Branch, Environment Canada, Dartmouth, Nova Scotia, Canada

Search for more papers by this author
Richard Bouchard

Richard Bouchard

National Data Buoy Center, NOAA, Stennis Space Center, Mississippi, USA

Search for more papers by this author
First published: 18 November 2011
Citations: 79


[1] Routine wave observations from buoys in the northeast Pacific now extend up to 35 years. Several recent studies reported long-term trends extracted from these records. However, significant modifications of the wave measurement hardware as well as the analysis procedures since the start of the observations result in inhomogeneities of the records. We analyze significant wave heights from seven offshore wave records. Several step changes of the mean monthly significant wave height of a few decimetres are identified. These changes are induced by buoy modifications and poor data quality rather than changes in the wave climate. After adjusting the data for these step changes the wave heights show positive trends for some of the southern locations and negative trends at the northern buoys, however all trends are much smaller than reported in previous studies. Storm wave heights are extracted from the occurrence rate distributions of the adjusted significant wave heights. No statistically significant trends can be established for storm wave heights.

Key Points

  • Buoy data do not show a consistent increase in average wave height
  • Operational data sets may contain artificial step changes
  • Data do not support an increase in storm intensities

1. Introduction

[2] Long term homogeneous wave measurements are important for the evaluation of trends and variability of the climate system. Of particular interest are trends in extreme values of the wave height due to their relevance to coastal management, marine transportation, fisheries, and marine exploration. On a global and basin-wide scale, wave climate studies are mainly based on hindcast products [Caires and Swail, 2004], on voluntary observing ship data [Gulev and Grigorieva, 2006], or on satellite observations [Young et al., 2011], whereas regional studies are often performed with buoy wave data. Operational wave measurements from buoys in the northeast Pacific began in the 1970s. Despite the shortcomings due to their fixed locations and irregular data gaps, they may provide information on short- to intermediate term fluctuations, and spatial patterns in the wave climate. Several recent studies address the variability of the NE Pacific wave climate with these buoy data sets [e.g., Allan and Komar, 2000; Gower, 2002; Menéndez et al., 2008; Ruggiero et al., 2010]. In general, these studies find increasing wave heights and increasing extreme wave heights, although the actual values fluctuate significantly, depending on the analysis techniques and the selection criteria for data acceptance.

[3] These buoys were designed for data input to operational weather and wave forecasts rather than monitoring long term trends. Buoys are serviced and their location rotated on regular service schedules, and the data analysis and the buoy hardware underwent several modifications. (See http://www.ndbc.noaa.gov/mooredbuoy.shtml for a description of the different buoy types). In addition, faulty sensors could generally affect data values for a period of time, if not eliminated in the quality control process. However, assessment and adjustment for data inhomogeneities have been limited. Such data adjustments are imperative before reliable results of a climate trend analysis are possible.

[4] Our analysis of the wave height measurements from seven offshore buoy stations (Figure 1a) shows that some of the buoy modifications resulted in inhomogeneities of the records. Using statistical tests to detect artificial step changes in the time series [Wang and Feng, 2009] theses shifts are corrected for and trends are estimated.

Details are in the caption following the image
(a) Buoy locations, and trends in daily mean values of significant wave heights. Heights are based on (b) unadjusted records Hs, and (c) on time series adjusted for non-climatic step changes Hadj, for entire record (dark grey), summer months (light gray), and winter months (white). Canadian buoys are identified with “C” in front of the WMO identifier.

2. Canadian Wave Buoys

[5] The Meteorological Service of Canada (MSC) has been operating the buoys C46036, C46184 and C46004 since the late 1980's. C46004 and C46036 were first installed and operated by the U.S. National Data Buoy Center (NDBC) and in 1987/88 transferred to MSC responsibility. Associated with this transfer were several significant hardware and processing changes (Table 1). (See the notation section for definitions of acronyms). The MSC wave data may be downloaded from http://www.meds-sdmm.dfo-mpo.gc.ca.

Table 1. Buoy history of payloads and hull types, and step changes Γ that occurred at the start of the new configurationa
Service Date Hull Payload Processor Γ [m]
46001 (Hav = 2.73m)
12/74–04/76 12D EEP EEP
07/76–09/79 10D PEB WSA −0.24
10/79–07/80 10D UDACS(A) WDA 0.22
08/80–06/82 10D GSBP WDA 0
07/82–03/90 6N GSBP WDA 0.19
07/90–05/06 6N DACT WA 0
06/06–current 6N ARES4.4 WPM −0.08
C46184 (Hav = 2.78m)
09/87–04/00 6N Zeno Zeno
04/00–current 6N WM WM
06/95–06/98 6N Zeno Zeno 0.16
05/99–04/00 6N01 Zeno861 Zeno861 −0.20
05/02–08/02 6N03 WM064 WM064 −0.29
09/02–05/03 6N03 WM064 WM064 −0.27
05/03–12/09 6N WM WM 0.41
46004/C46004 (Hav = 2.87m)
10/76–10/77 10D PEB WSA
09/78–05/80 12D PEB UDACS WSA 0
02/81–06/83 12D UDACS(A) WDA 0.32
06/83–06/88 6N GSBP WDA 0.12
06/88–05/99 6N Zeno Zeno 0
05/99–02/00 6N05 WM036 WM036 −0.20
05/01–04/02 6N03 WM064 WM064 −0.66
05/02–12/10 6N WM WM 0.59
46036/C46036 (Hav = 2.87m)
08/86–09/87 6N GSBP WDA
09/87–07/98 6N Zeno Zeno
07/98–current 6N WM WM
07/89–06/94 6N Zeno Zeno 0.17
09/94–07/96 6N02 Zeno862 Zeno862 −0.28
04/00–05/01 6N03 WM064 WM064 −0.09
05/06–11/07 6N02 WM092 WM092 0.13
11/07–05/09 6N03 WM039 WM039 −0.20
05/09–05/10 6N01 WM156 WM156 0.22
46005 (Hav = 2.77m)
10/76–08/78 10D PEB WSA
08/78–06/80 12D PEB WSA 0
06/80–11/81 6N MXVII(MOD) WDA 0.27
11/81–10/85 12D UDACS(A) WDA 0
02/86–08/90 6N GSBP WDA 0.21
01/91–09/04 6N DACT WA 0
09/04–06/10 6N ARES WPM −0.14
06/10–current 3D AMPS WPM 0
46002 (Hav = 2.71m)
07/75–05/79 10D PEB WSA
05/79–09/80 10D UDACS(A) WDA 0.17
09/80–12/82 10D GSBP WDA 0
01/83–02/90 6N GSBP WDA 0.21
06/90–06/06 6N DACT WA 0
06/06–07/09 6N ARES WPM −0.11
46006 (Hav = 2.82m)
04/77–08/79 10D PEB WSA
08/79–07/85 12D UDACS(A) WDA 0.25
08/85–05/86 6N GSBP WDA 0
06/86–03/88 12D GSBP WDA 0
08/88–01/91 12D DACT WA 0
06/92–07/00 6N DACT WA 0.13
08/00–08/07 6N VEEP WA 0
08/07–08/08 6N DACT WA 0
08/08–current 3D ARES WPM −0.13
  • a Shown are the start and end days of the deployment [mm/yy], the hull type (12m Discus buoy, 12D, 10m Discus buoy, 10D, 6m NOMAD buoy, 6N, or 3m Discus buoy, 3D), payload type, wave processor type, and step size. Note, only buoy modifications associated with Hs changes are listed for MSC buoys. Hav is the mean value over the entire record length.

[6] The buoys record vertical acceleration at 1 Hz sampling rate for 34 minutes every hour. Double integration of the accelerations yields the surface elevation η, and the significant wave height Hs = 4σ(η), where the standard deviation σ(η) is inferred from the power spectrum of the surface elevation [AXYS, 1996]. The low frequency cut-off is 0.03Hz, and no high-frequency noise filtering is applied. Thus, Hs is not a directly observed quantity but is calculated onboard. Two different onboard processors have been used, the Zeno until the 1990s, and the Watchman which replaced it (Table 1).

[7] All three buoys were originally equipped with a gimballed Datawell wave sensor. In 1998–2000 these vertically stabilized accelerometers were replaced with single-axis strapped-down accelerometers (SDA).

[8] Direct comparisons revealed that SDA's on 6m NOMAD buoys (6N) under-report wave heights by 2% to 10%, depending on sea states, compared to Datawell buoys [Skey et al., 1999]. In contrast, Bender et al. [2010] find that a 3m discus buoy (3D) equipped with SDA overestimated the true wave height by 26% during the passage of a hurricane. An ongoing inter-comparison between Datawell Waverider, equipped with vertically stabilized wave sensors, and two MSC buoys (3D and 6N) with SDA aims at resolving this issue (www.jcomm.info/WET).

3. U.S. Wave Buoys

[9] The NDBC buoys 46001, 46002, 46005 and 46006 were deployed in the 1970's and hull forms and payload were changed several times since then (Table 1). The reported hourly Hs values were calculated from 20 minute observations, however data from the current family of wave processors, WPM and its derivatives, are based on either 20 or 40 minute samples. In the beginning (PEB-type payload) the data rate was every 3 hours. Data may be downloaded from http://www.ndbc.noaa.gov/.

[10] The processing of NDBC wave buoy data underwent several major modifications. In 1977 an empirical, constant low frequency noise correction was introduced, and in 1984 updated to be a function of the sea-state, and further improved in 1987 [Lang, 1987]. Initially, the height range of individual waves was limited to ±11m, but was increased to ±20m with the introduction of GSBP. The upgrade from WSA to WDA increased the number of spectral bands from 12 to 35. Low frequency resolution was further increased with the WPM processor. The early PEB and UDACS peak significant wave heights were only 7 to 9 m, with higher peak values of Hs being reported since the change to the UDACS(A). It is likely that the coarse frequency resolution and the limited individual wave height range resulted in capping the maximum significant wave height.

[11] All NDBC wave sensors are SDA types. Additional information on NDBC wave processing changes is contained in Steele et al. [1992] and Earle [1996].

4. Non-climatic Step Changes in Hs

[12] The records of the significant wave height Hs and its monthly means equation image are clearly dominated by seasonal modulations (Figure 2a), with smaller multi-year fluctuations, particularly in the winter wave heights. While El Niño and La Niña events often resulted in increased wave heights, anomalous wave heights are not unique to these events. To analyze whether step-like, long term changes in the record occurred, we use the software package RHTestsV3 [Wang and Feng, 2009] to perform a homogeneity test on the equation image time series to identify potential change points. This tool has been used successfully for, e.g., the homogenization of wind speeds [Wan et al., 2010] or individual wave buoy records [Thomas and Swail, 2009].

Details are in the caption following the image
Extraction of step changes and trends for 46005. (a) Hourly Hs observations (black) and calculated monthly mean values equation image (red). (b) Difference between observations and GROW2000 reference series equation image (black), and step changes identified using RHTestsV3 (blue line). (c) Monthly mean values (black) and trend lines: for raw monthly input data (red) and multi-segment data (blue).

[13] Here we analyze the mean monthly wave heights from all seven stations, restricting the equation image data to months with ≥60% data coverage (Figure 2a). To increase the sensitivity and reliability of the detection of change points, the GROW2000 hindcast data Href at the nearest grid point [Oceanweather, 2007] are supplied as additional input to RHtestsV3. These reference time series are representative of the climate trends and low-frequency variability of the target equation image series. In this configuration, the penalized maximal t test is used to detect unknown change points from the difference time series equation image = equation imageHref (Figure 2b), and the Student t test for testing the significance of known change points (see, e.g., Wang [2008] for references).

[14] Statistically significant step changes Γ occurred at all seven stations (Table 1). In nearly all cases these changes may be directly associated with known buoy modifications, or poor data quality. For example, C46004 reported too low in 05/2001–05/2002. This fault remained unnoticed and the buoy and payload were redeployed as C46184 from 05/2002 to 05/2003. These two deployments caused the largest step changes with Γ = O(±0.6m).

[15] For the NDBC buoys a general pattern emerges with i) a positive shift at the discontinuation of the early PEB buoys, ii) a small positive step at the hull change from the 10m or 12m Discus buoys to the 6m Nomad buoys, and iii) a small negative step at the change to the most recent processing system (WPM). Step changes at the MSC buoys are less consistent.

5. Time Series Adjustment and Long Term Trends in Hs

[16] Previous studies did not account for the non-climatic shifts identified above, likely introducing errors to the trend values of the Hs records. Using RHTestsV3, the equation image time series is partitioned along the detected change points and a multi-phase trend line with a common slope is calculated (Figure 2c). At all seven stations, this trend is weaker than the trend obtained from the raw equation image time series (Figure 2c, only 46005 shown).

[17] The homogeneity tests are applied to the monthly mean equation image series. The original input time series of hourly Hs data is then adjusted to the most recent segment, based on relative step changes:
equation image
where 〈〉 denotes the mean value, and ti, M are the time and the total number of step changes, respectively. This adjustment scheme assumes that buoy modifications mainly affected the sensor gain rather than being a fixed offset, and uncertainties are largest for extreme wave heights.

[18] Here we estimate the trends from the daily mean values of the hourly observations equation image up to 31/03/2010. These trends are less affected by irregular data coverage than the monthly averaged data. The data are also divided into summer (April 1–Sep 30) and winter (Oct 1–Mar 31) observations to reduce the effect of seasonal fluctuations. Trends and their errors are calculated with the generalized least squares (GLS) method where the autocorrelated errors are estimated based on the AR(1) model.

[19] The trend values of the unadjusted observations equation images range from −23mm/a to +22mm/a, and the sign of the trend is divided along US and Canadian buoys, rather than geographically (Figure 1b). These apparent trends would suggest a noticeable effect on the wave height over the last 2–3 decades, and have been associated with climate change [Ruggiero et al., 2010] (based on 46005 and 46002 data).

[20] However, these trends are to some extent introduced by modifications of the measurement techniques (hardware and software) and poor data quality. Trends estimated from the daily means of adjusted hourly data equation imageadj are between −9mm/a and 8mm/a, and are positive at the southernmost station, and negative at the northern locations. At the central locations only the winter data at 46005 show a statistically significant trend (+8mm/a). Trends of the adjusted data are independent of ownership of the buoys, whereas the trends of the raw data showed a sharp jump at neighbouring buoys operated by different agencies. (Figures 1c and 1b). There is some indication that the wave heights during the winter months show less of a trend than the wave heights during summer (except at 46005).

6. Storm Wave Heights

[21] Of particular interest are the wave heights during storms. Ruggiero et al. [2010] report that the significant wave height during storms Hsevere (which they defined as the average of the five highest independent significant wave heights within 1 year) at station 46005 increased by 71 mm/a, compared to 15 mm/a they find as trend of all Hs data. Such a rapid increase in storm wave heights would have significant impact on coastal erosion and marine operations. However, the analysis by Ruggiero et al. [2010] did not correct for any artificial shifts. In particular, the restricted PEB data at the beginning of the records, when maximum wave amplitudes were limited to 11m, may generate erroneous trends in extreme wave heights. Limiting the Hsevere trend analysis to the raw data after 06/1980 yields no trend that is statistically different from zero. Similarly, using the adjusted data and the same criteria for data inclusion as Ruggiero et al. [2010], we find no statistically significant trend in Hsevere at this location.

[22] Here we take a different approach at estimating average severe wave heights. The peak significant wave height of a single storm can be defined as
equation image
with storm duration N = 24 (corresponding to 24h). Thus, FP(equation images(t) < z) is the probability that the significant wave height during a storm does not exceed a value z, and the exceedance probability is P(equation imagesz) = 1 − F1/N (for references see Gemmrich and Garrett [2011]). While wind speeds over the ocean are often described by a Weibull distribution (for references see Monahan [2006]) there is no theoretical model for the distribution of significant wave heights, although the Weibull distribution, amongst others, has been suggested. Figure 3a shows examples of probability distributions of storm wave heights equation images obtained from 5 year wave record segments. The body of the distribution is well represented by a Weibull distribution. However, the largest storm wave heights are less frequent than the Weibull fit would predict.
Details are in the caption following the image
(a) Probability distributions of maximum significant wave heights during storms, observed at C46036 during 2001–2005 (black) and 2006–2010 (gray). The straight dashed lines depict Weibull distributions fitted to the data. The horizontal dotted lines represent probabilities of (top) P = 2.3 × 10−4 and (bottom) P = 6 × 10−3, corresponding to 2 and 52 occurrences per year, respectively. (b–h) Average storm significant wave heights equation imagex occurring with a probability of 2/year (triangles) and 52/year (circles). Dashed lines show least square fits calculated from periods when the length of summer and winter data differed by less than 20% (solid symbols). The slopes and their uncertainties are given.

[23] Average wave heights of severe storms may be described as the height of storms occurring with a probability of, say, twice per year (P = 2.3 × 10−4). For comparison we take storms that occur with a probability P = 6 × 10−3 (i.e. 52 times per year) as a background, and the associated wave height represents moderate conditions (Figure 3a). None of the trends in storm wave heights (Figures 3b3h) are statistically different from zero.

7. Conclusion

[24] The operational wave observations from buoys in the northeast Pacific have been in place for up to 35 years, but have been subject to significant modifications that resulted in several artificial step changes of the long term record of significant wave heights. The replacement of the PEB/WSA-type payload by more advanced systems (about 1980) introduced the most consistent and significant step in all affected buoys. Hull changes and further modifications of the payload and analyzing systems resulted in less significant changes. The largest step changes are associated with temporary faults in the equipment.

[25] Accounting for these changes, trends for the corrected data are substantially smaller than the apparent trends obtained from the uncorrected data, and a regional pattern, independent of wave buoy type, emerges. A small decrease in wave heights exists for the region off Alaska and British Columbia, whereas trends off Washington and Oregon are not statistically significant (except winter data at 46005). Wave heights off northern California (annual and summer months only) have slightly increased. The strongest increase occurred for winter wave heights off Washington (8mm/a). This pattern is in accordance with a recent coarse grid global estimate of wave height trends [Young et al., 2011]. We estimate the wave height of severe storms from the distribution of independent storm wave heights and find that no increase in wave heights during severe storms occurred. This is in contrast to a previous study [Ruggiero et al., 2010] that reported substantial increase of storm wave heights at stations 46002 and 46005, but is based on unadjusted data. The wave climate fluctuates at short terms of a few year period, and no coherent decadal trends emerge. The wave records need to be extended for many more years before long term wave height trends can be established with certainty.

[26] The analysis of trends and variability of wave climates is an ongoing process. Therefore, it is important that adjusted wave data are routinely made available in the historical wave data archives of the National Oceanic and Atmospheric Administration (NOAA) and the Integrated Science Data Management (ISDM), as well as in the International Comprehensive Ocean Atmosphere Data Set (ICOADS).


  • Payload
  • data logger, including CPU and firmware to control the sensor and compute averages. Does not include the wave sensors.
  • Wave processor
  • software to process raw wave data
  • EEP
  • Engineering Evaluation Phase
  • PEB
  • Prototype Environmental Buoy
  • UHF Data Acquisition and Control System
  • UDACS(A)
  • UDACS, using WDA
  • GSBP
  • General Service Buoy Payload
  • DACT
  • Data Acquisition and Control Telemetry
  • VEEP
  • Value Engineered Environmental Payload
  • ARES
  • Acquisition and Reporting Environmental System
  • AMPS
  • Advanced Modular Payload System
  • WSA
  • Wave Spectrum Analyzer (12 frequency bands)
  • WDA
  • Wave Data Analyzer
  • WA
  • Wave Analyzer
  • WPM
  • Wave Processing Module
  • WM
  • Watchman
  • Navy Oceanographic Meteorological Automatic Device
  • Acknowledgments

    [27] K. Steele, E. Michelena (formerly NDBC), and V. Williams (Environment Canada) provided information on early NDBC systems, and the MSC buoys, respectively. X. Wang (Environment Canada) advised on using the RHtestsV3 software package and provided an internal review. Comments by V. Swail (Environment Canada) further helped to improve the manuscript. M. Müller (University of Victoria) provided the GLS trend calculations. A review by P. Bromirski helped to improve the clarity of the text. This study was supported by funds of Canada's Natural Sciences and Engineering Research Council.

    [28] The Editor thanks Peter Bromirski and an anonymous reviewer for their assistance in evaluating this paper.